A stock forecast model using variable selection by real-value GA

Abstract

One of the major challenges of using big data to predict stock prices is the prevention of overfitting. In this study, we observe how the overfitting problems of the stock price forecast model is mitigated by adopting the variance of a parameter with which the evaluation score is the best that was generated from the repetition of generation changes in the real-value GA. In any method of machine learning used in the research (deep learning, gradient boosting, SVM), the model accuracy was improved when variable selection by the variance of a parameter was added.